Population genetic analysis reveals a geographically limited transition zone between two genetically distinct Atlantic salmon lineages in Norway

Abstract Atlantic salmon is characterized by a high degree of population genetic structure throughout its native range. However, while populations inhabiting rivers in Norway and Russia make up a significant proportion of salmon in the Atlantic, thus far, genetic studies in this region have only encompassed low to modest numbers of populations. Here, we provide the first “in‐depth” investigation of population genetic structuring in the species in this region. Analysis of 18 microsatellites on >9,000 fish from 115 rivers revealed highly significant population genetic structure throughout, following a hierarchical pattern. The highest and clearest level of division separated populations north and south of the Lofoten region in northern Norway. In this region, only a few populations displayed intermediate genetic profiles, strongly indicating a geographically limited transition zone. This was further supported by a dedicated cline analysis. Population genetic structure was also characterized by a pattern of isolation by distance. A decline in overall genetic diversity was observed from the south to the north, and two of the microsatellites showed a clear decrease in number of alleles across the observed transition zone. Together, these analyses support results from previous studies, that salmon in Norway originate from two main genetic lineages, one from the Barents–White Sea refugium that recolonized northern Norwegian and adjacent Russian rivers, and one from the eastern Atlantic that recolonized the rest of Norway. Furthermore, our results indicate that local conditions in the limited geographic transition zone between the two observed lineages, characterized by open coastline with no obvious barriers to gene flow, are strong enough to maintain the genetic differentiation between them.


| INTRODUC TI ON
Sustainable management of biodiversity in exploited species requires among other things, an understanding of their structuring into distinct breeding populations, as well as the nature and extent of population connectivity and adaptive population differentiation.
Elucidating connectivity among populations, and identifying the underlying mechanisms that shape observed patterns, represents an ongoing challenge. Given the ever-increasing pressure on much of the world's biota and ecosystems, this is increasingly urgent. For the Atlantic salmon (Salmo salar L.), an iconic and economically important anadromous fish that has and continues to be subjected to a diverse array of anthropogenic challenges (Forseth et al., 2017;Glover et al., 2017;Parrish, Behnke, Gephard, McCormick, & Reeves, 1998;Taranger et al., 2015), it has never been more important to map populations, and quantify their evolutionary and contemporary relatedness and connectivity.
Atlantic salmon inhabit cold-water rivers on both sides of the north Atlantic. In anadromous populations, the quintessential form, fertilized eggs are deposited in well-oxygenated gravel areas, and after hatching, juveniles spend 1-5 + years in freshwater before migrating to the sea (Klemetsen et al., 2003;Metcalfe & Thorpe, 1990). After 1-3 + years of oceanic feeding, they mature and return to freshwater to reproduce, completing the life cycle. The species' anadromous life history involves long-distance migrations from individual spawning rivers and tributaries to shared oceanic feeding areas where fish from multiple populations and regions meet (Bradbury et al., 2016;Gilbey et al., 2017;Olafsson et al., 2016;Sheehan, Legault, King, & Spidle, 2010), with all but a very small fraction of returning salmon, homing back to their natal rivers (Jonsson, Jonsson, & Hansen, 2003;Stabell, 1984). Accurate homing and fidelity to natal river provides the isolating mechanism through which genetically distinct populations have been able to establish in this species throughout its native range (Bourret et al., 2013;King, Kalinowski, Schill, Spidle, & Lubinski, 2001;Ståhl, 1987;Verspoor et al., 2005). In turn, this has also provided the basis for the evolution of genetic differences in life-history traits among populations, some of which may be adaptive (Garcia de Leaniz et al., 2007;Taylor, 1991).
Population genetic structure in Atlantic salmon is often, but not always, associated with isolation by distance (Dillane et al., 2007;Glover et al., 2012;Perrier et al., 2011;Primmer et al., 2006). To some extent, this will be because the level of contemporary straying among populations is a function of distance. However, other factors such as landscape features (Dillane et al., 2008), association with climate clines through local adaptation (Gilbey, Verspoor, & Summers, 1999;Jeffery et al., 2017;Verspoor, Fraser, & Youngson, 1991), and colonization history in connection with ice-cap retreat patterns (Cauwelier et al., 2018;Olafsson et al., 2014;Rougemont & Bernatchez, 2018), play an important role in shaping population genetic structure in this species. Other factors may well be involved and all are likely to be of variable importance in defining levels of within and among river population differentiation.
Norway and Russia have approximately 400 and 110 rivers containing Atlantic salmon populations, respectively (http://www. nasco.int/River sData base.aspx) and populations in this region represent a large proportion of the wild Atlantic salmon resources globally. Yet, despite the significance of this region for Atlantic salmon, a detailed picture of population genetic structure in Norway is lacking, with the literature on Norwegian rivers confined largely to scattered population samples within broader scale assessments (Bourret et al., 2013;Verspoor, 1997;Wennevik et al., 2004), although some Norway-specific population genetic studies have also been published (Glover et al., , 2012. Russian populations have been more extensively studied. Early studies using allozyme (Kazakov & Titov, 1991) and mitochondrial DNA markers (Makhrov, Verspoor, Artamonova, & O'Sullivan, 2005) described some of the major structuring of Atlantic salmon populations of the Russian north. However, several more recent studies, applying different classes of markers, have extended understanding of the population structure and the recolonization history of these northern populations since the last glaciation. Asplund et al. (2004) Sea and an eastern group. In a study of Atlantic salmon populations from the Baltic, White and Barents Seas, Tonteri et al. (2005) concluded that it was most likely that the populations from the White and Barents Seas were colonized from multiple refugia, one possibly located in the eastern Barents Sea. In a follow-up study with populations from the White and Barents Seas, Tonteri et al. (2009) found evidence of four distinct population clusters; Atlantic Ocean and western Barents Sea, Kola Peninsula, western White Sea and eastern Barents Sea. More recently, Ozerov et al. (2017) developed a high-density genetic baseline for northern Atlantic salmon populations, and also briefly described population structure, identifying seven major population complexes, largely consistent with the results from the above-mentioned studies.
The primary objective of the present study it is to provide the first detailed analysis of the population genetic structure of salmon stocks across the whole of Norway and western Russia. This analysis encompasses data for 9,165 salmon from 115 rivers analyzed for a panel of 18 microsatellite DNA markers. The secondary objective of this study is to place the data set in the public domain to facilitate comparative and integrated analyses of structuring patterns across the species' range.

| Sampling
In total, 9,165 individuals were sampled in 115 rivers from the Komi Republic in Russia to the Østfold region in southern Norway (Figure 1). This included samples of individuals from different stages of their life cycle (parr, fry, smolt, and adult), although most were of juveniles (fry & parr) collected by electrofishing at 2-4 locations within each river. In all cases, sampling encompassed individuals representing all juvenile year classes present at that particular sampling location. Fish were euthanized using an overdose of benzocaine, and fin clips were taken and transferred to tubes with 96% ethanol. Permits for collection of the samples were issued by County Governors in Norway, and by the Federal Agency for Fisheries in Russia. For simplicity, river samples are referred to as "population samples."
Data were screened using the software COLONY ver. 2.0.5.1 (Jones & Wang, 2010), which implements full-pedigree likelihood methods to simultaneously infer sibship and parentage among F I G U R E 1 Map showing the location of the rivers sampled in Russia and Norway. Numbers refer to river names in Table A1. The major genetic division of the populations into two groups are indicated with a dashed line individuals using multilocus genotype data, to purge the data set from full siblings that would lead to bias in allele frequency estimates as suggested by Allendorf and Phelps (1981), but see work by Waples and Anderson (2017). Analyses were run with no information on parental genotypes, assuming both male and female polygamy as well as possible inbreeding. The full-likelihood model was chosen together with run length and precision set to medium. A total of 1,007 individuals were removed (Table A1).

| Statistical analysis
Screening for outlier loci was performed using two methods. First, with the approach implemented in ARLEQUIN v.3.5.1.2 (Excoffier, Laval, & Schneider, 2005), which accounts for historical meta-population structure with a hierarchical island model (H) (Excoffier, Hofer, & Foll, 2009) thus aiming to reduce the number of false positive F ST outlier loci. The underlying assumptions are that the average migration rate between populations on different islands is lower than that between demes on the same island and that the heterozygosity between populations can be inferred using the heterozygosity within a population (Excoffier & Lischer, 2010). Significance of outliers was assessed by running 50,000 simulations, 100 demes, and 20 groups. Second, with the Fdist approach (Beaumont & Nichols, 1996) implemented in LOSITAN (Antao, Lopes, Lopes, Beja-Pereira, & Luikart, 2008) in which loci with an unusually high F ST are considered to be putatively under directional selection. We simulated the neutral distribution of F ST with 1,000,000 iterations at a significance level of 0.001 under a stepwise mutation model. This method also implements a multi-test correction based on false discovery rates (FDR) to avoid high overestimation of the percentage of outliers (e.g., 1% of false positive with a threshold of 99%). Due to the impossibility of handling data sets exceeding 100 populations, LOSITAN was conducted separately for each of the regional divisions obtained from STRUCTURE.
Total number of alleles and allelic richness (A r ) were calculated with MSA (Dieringer & Schlötterer, 2003), whereas observed (H o ) and unbiased expected heterozygosity (uH e ) were computed with GenAlEx (Peakall & Smouse, 2006). The genotype distribution of each locus per year class and its direction (heterozygote deficit or excess) was compared with the expected Hardy-Weinberg distribution using the program GENEPOP 7 (Rousset, 2008) as was the linkage disequilibrium. Both were examined using the following Markov chain parameters: 10,000 steps of dememorization, 1,000 batches and 10,000 iterations per batch. Significance was assessed after applying sequential Bonferroni correction (Holm, 1979). Effective population size (Ne) based on linkage disequilibrium was estimated using LDNe v1.31 (Waples & Do, 2008) using the random mating option and the Pcrit = 0.02 criterion for screening out rare alleles, and with 95% confidence intervals derived from a jack-knife approach.
Allelic richness and heterozygosity were tested for latitudinal trends using the nonparametric Kendall measure of rank correlation (Kendall & Gibbons, 1976), which measures the similarity of the orderings of the data when ranked by north-south gradient or by the value of the variable tested (Valz & Thompson, 1994), and implemented in the R Package "Kendall" (R Core Team, 2016).
Besides, conservation limits (i.e., the number of spawning salmon needed for fully exploiting the rivers potential for production of juveniles) expressed as kg of female fish were tested for correlation with three different variables: N e , H o , and A r . River-specific conservation limits information was only available for the Norwegian rivers.
Potential recent declines in effective population size were assessed using the software BOTTLENECK v1.2.02 (Piry, Luikart, & Cornuet, 1999) based on allele frequencies. As the data set was genotyped at <20 microsatellites, Wilcoxon's test and the graphical mode shift indicator were chosen (Piry et al., 1999). Likewise, loci were assumed to evolve under the two-phase mutation model (Di Rienzo et al., 1994) with 5% of the mutations involving multiple steps with a variance of 12 (see Tonteri et al., 2009). Statistical significance of the Wilcoxon's test was assessed by 2,000 replications followed by the sequential Bonferroni correction for multiple significance tests.
Hierarchical population structure was explored using STRUCTURE (Pritchard, Stephens, & Donnelly, 2000) and traditional F ST (Weir & Cockerham, 1984). STRUCTURE v. 2.3.4 was used to identify genetic groups under a model assuming admixture and correlated allele frequencies using population information to assist the analysis. STRUCTURE was analyzed following a hierarchical approach (Gilbey et al., 2017) using the program ParallelStructure ) that distributes jobs between parallel processors in order to significantly speed up the analysis time. Ten runs with a burn-in period consisting of 250,000 replications and a run length of 750,000 MCMC iterations were performed for K = 1 to K = 20 clusters for the total data set. To determine the number of clusters in which samples could be divided into, the STRUCTURE output was analyzed by combining the visual inspection of the barplots with the ad hoc summary statistic ΔK of Evanno, Regnaut, and Goudet (2005), which is based on the rate of change of the "estimated likelihood" between successive K values and allows the determination of the uppermost hierarchical level of structure in the data.
The data set was split into smaller units based upon this analysis until coherence in the clusters were lost, or until single rivers appeared as independent entities. Finally, runs for the selected Ks were averaged with CLUMPP v.1.1.1 (Jakobsson & Rosenberg, 2007) using the LargeK Greedy algorithm and the G' pairwise matrix similarity statistic and were graphically displayed using barplots. STRUCTURE allowed the partitioning of the data set into subsets of geographic regions that were analyzed in a hierarchical manner.
The analysis was performed on populations (i.e., merged river samples) using a covariance matrix with 10,000 permutations. The results from the analysis were visualized as plots constructed in Microsoft Excel. The relationships among genetic distance and geographical distances were examined via a simple Mantel (1967) test between the matrices of pairwise F ST and geographical distance. Mantel tests were conducted with PASSaGE (Rosenberg & Anderson, 2011), and significance was tested after 10,000 permutations. The program PGDSpider 2.1.1.3 (Lischer & Excoffier, 2012) was used to conduct the file conversion to the software used for the different analyses when required.
In order to further investigate the geographically limited transition zone identified by STRUCTURE (see results), we conducted a cline analysis to estimate the shape, center, and width of the cline generated by our molecular data (Gay, Crochet, Bell, & Lenormand, 2008).
Geographic cline analysis over a 3,600 km transect starting in Unya in the Komi Republic in Russia to Enningdalselva in the Østfold region in the Norwegian-Swedish border were conducted with the R package HZAR (Derryberry, Derryberry, Maley, & Brumfield, 2014). The 15 models implemented in HZAR were fitted to the normalized loading of the first principal component analysis (PCA) axis based both on the panel of 18 microsatellites as well as on each locus independently to determine the position, width, and shape of clines over the total geographic distance. The reference cline was built using STRUCTURE Q-score for the total data set and, in both cases, the best cline model was decided upon AIC scores. Clines were considered significantly displaced if the two log-likelihood unit support limits of the cline center did not overlap with the STRUCTURE Q-score (Qb = 1−Qs).

| RE SULTS
The raw genetic data for all of the individuals included in the present study are deposited in Appendix S1.

| Genetic variation within populations
ARLEQUIN reported two outlier loci (Ssa289 and MHC2) in the full data set, whereas LOSITAN suggested that MHC2 was the only locus under directional selection in the two main clusters resulting after the first hierarchical division of the 115 samples. Thus, using a combined approach, MHC2 remained the only candidate for directional selection. The influence of this locus was tested by conducting STRUCTURE with and without it (Appendix S2). As inclusion/exclusion of this locus had no influence on the resulting genetic structure, MHC2 was retained in all the subsequent analyses.
Hardy-Weinberg deviations were reported in ~10% of the tests performed across populations for every locus, but they were reduced to 2.9% after sequential Bonferroni correction. Likewise, the percentage of deviations from LD decreased from 16.5% to 5% after correction. In both cases, the departures from expectations were distributed across populations and loci, therefore, no loci were dropped from the data set based on the results from these analyses.
Over the 18 microsatellites, a total of 413 alleles were observed, ranging from 5 to 7 alleles in SsaD486 and Ssa14, respectively, to 41 in SsaD144 and SsaD157. The total number of alleles per population ranged from 78 to 259, with a mean of 217 (Table A1). The Kovda (3) river showed an extremely low number of alleles: 78 in 26 individuals whereas, for example, 190 alleles were reported from 24 individuals sampled in the river Soknedalselva(109). The average number of alleles per locus within a population ranged from 4.3 in Kovda (3) to 14.4 in Eidselva(82), whereas allelic richness ranged from 4.3 in Kovda(3) to 10.76 in Otra(112).
In Norway, the conservation limits expressed as kg of female fish per river were significantly correlated with Ne (r 2 = 0.2085, p < 0.001), but not with Ho (r 2 = −0.01, p = 0.68) nor with Ar (r 2 = 0.004, p = 0.25). After performing Bonferroni correction for multiple comparisons, the Wilcoxon test did not reveal any population displaying evidence of having experienced recent bottlenecks.
Likewise, no mode shift in allele frequencies was detected in any of them, all showing L-shaped allele frequency distributions; that is, the number of alleles in the low-frequency classes (<0.1) exceeded the number of alleles in the higher frequency ones.

| Among-population genetic structure
The first hierarchical level of division detected by ΔK test of STRUCTURE results showed two clusters (ΔK = 176.5) that divided the data set in a northernmost group ranging from the rivers Unya(1) to Reisa(51) (i.e., 51 sampled rivers), and a southern cluster from the rivers Laukhelle(53) to Enningdalselva(115) (63 rivers) ( Figure 1). The ancestry of the population in the river Målselva (52) was almost evenly split between both clusters. At the second hierarchical level of division, further structure was revealed among populations ( Figure 3). In the northern group, the eastern populations from Unya(1) to Kitsa(8) formed a distinct cluster in the plots from the Structure analysis, different from the populations draining to the Barents Sea coast on the northern side of the Kola Penisula.
The river Ponoi(9) appears as a transitional river. This genetic division also corresponds to a change in life history as the eastern populations and the White Sea populations are mainly "autumn-run" salmon, which ascend the river the more than a year before spawning, while the Barents Sea rivers are dominated by "summer-run" salmon spawning in the same year they return to the river. On the northern coast of the Kola Peninsula, there seems to be a genetic shift between the rivers east (10-20) and west 26-36 of the Kola Bay ( Figure 3b). The rivers draining into the freshwater Tuloma lake (21-24) form a distinct cluster. Another genetic shift can be observed between rivers Bergebyelva(36) and Vestre Jakobselv (37) in the inner part of the Varanger Fjord. The two Tana tributaries Iesjohka(43) and Laksjohka (44)

| Investigation of the transition zone by cline analysis
The PCA cline based on the total 18 microsatellites fitted a fixB model, with the center situated at 1,621 km from the Unya(1) and with a width of 296 km ( Figure 6, Table S1-Appendix S2). Both the center and the width of this cline were geographically located between F I G U R E 3 Hierarchical Bayesian clustering of the 115 populations using locprior information in Structure F I G U R E 2 Patterns of genetic diversity assessed along the coastline from Russia to southern Norway. Genetic diversity was measured as: (a) Ho and overall allelic richness (Ar) per population; and (b) Ar for two out of the eleven loci (Sp1605, SsaF43, MHC1, SsaD486, SSspG7, Ssa14, Ssa289, MHC2, SsaD157, SSsp2210, Ssa197) showing significant trends. (c) Ar for MHC2 and SS202 (the second one showing no significant trend, for the sake of the comparison). Both the overall Ar for the 18 loci (τ = −0.281, p = 9.37e−06) and Ho (τ = −0.255, p < 0.001) experienced a significant decline from south to north. The vertical dashed line shows the first level of STRUCTURE division of the data set the rivers Reisa(51) and Målselva(52), in very close agreement with the results from STRUCTURE (Figure 3). The PCA cline overlapped with the STRUCTURE Q-score cline, which also met a fixB model, with the center located at 1,600.6 km (also between rivers Reisa (51) and Målselva(52)) and 336.4 km of width. The clines generated by the microsatellite loci SSsp2210, SSspG7, SsaD144, MHC1, Ssa197, Sp2216, MHC2, SsaF43, and Ssa202 presented their centers within the width of the reference cline based on the STRUCTURE Q-score.
Loci SsaD486 and SSsp2201 showed clines centered further south, unlike loci Ssa289, SSsp3016, SsaD157, Ssa14, Sp1605, SsOsl85, and Ssa171, which were centered between the rivers Kovda(3) and Tana-Iesjohka(43). The graphical representation of the clines computed for each marker separately is shown in Appendix S3- Figure S1. Based on the main observations detailed above, we conclude that Atlantic salmon in Norway originate mainly from two genetic lineages, one from the Barents-White Sea refugium that recolonized northern Norwegian and adjacent Russian rivers, and one from the eastern Atlantic that recolonized the rest of Norway. We also conclude that local conditions in the geographically limited transition zone between these two lineages in northern Norway, characterized by a relatively open coastline with no obvious barriers to straying nor gene flow, are strong enough to maintain its character since its post-last glacial maximum establishment.

| D ISCUSS I ON
Whether or not selection, restricted straying and gene flow, or other mechanisms are responsible for its maintenance, remains to be elucidated.

| Phylogeographic patterns in northern Norway and northwest Russia
The distinctiveness of salmon in northern Norway and northwest report a more precise geographic location of the clear genetic break in Norway, potentially reflecting the recolonization ranges from different lineages, was a study of microsatellite genetic variation in 21 Norwegian populations (Glover et al., 2012). These authors identified a genetic division in the geographic region between Målselva and Roksdalsvassdraget (populations 52 and 54 in the present study) which is consistent with the division revealed from the analysis here (Figure 3). Subsequent studies with SNPs, primarily aimed at investigating introgression of domesticated Atlantic salmon escapees in Norwegian populations, have also detected a distinct genetic change in this region Karlsson, Diserud, Fiske, & Hindar, 2016).
The existence of a clear genetic divide in northern Norway is most likely to reflect the postglacial colonization history of this region, and the influence of mechanisms maintaining this divide over time. As mentioned above and in the introduction, several studies have suggested that the northern areas of Russia and Norway were colonized by different lineages, originating from different refugia.
The eastern part of the Barents Sea was not entirely covered by ice during the last glacial maximum (Hughes, Gyllencreutz, Lohne, Mangerud, & Svendsen, 2016), and this area has been suggested as the location of a refugium from which the northeastern part of the distribution range of Atlantic salmon was colonized (Asplund et al., 2004;Kazakov & Titov, 1991;Nilsson et al., 2001;Rougemont & Bernatchez, 2018;Tonteri et al., 2005Tonteri et al., , 2009. Asplund et al. (2004) suggested that populations east of the genetic divide, observed in the eastern part of the Kola peninsula (this divide also shown by Tonteri et al., 2009 and present in our data set), primarily originated from this eastern refugium, while populations on the northern side of the peninsula and westwards into northern Norway mainly originated from other Atlantic lineages. Our data are consistent with this.

Several studies have demonstrated the presence of North
American alleles/haplotypes in populations along the Barents Sea coast (Asplund et al., 2004;Bourke et al., 1997;Makhrov et al., 2005;Nilsson et al., 2001;Rougemont & Bernatchez, 2018) suggesting a contribution from both eastern and western Atlantic lineages. Based on a joint analysis of both Esterase-D* and mtDNA, Mahkrov, and colleagues (Makhrov et al., 2005) first proposed that the genetic affinities of the region's salmon populations to those in North America arose from the unique postglacial recolonization of the area by salmon from both Europe and North America. In combination with other observations (Asplund et al., 2004;Tonteri et al., 2009)

| The geographically sharp transition zone between the eastern Atlantic and Barents-White Sea lineages
The continued existence of a geographically limited transition zone in northern Norway between two highly divergent regional salmon lineages raises both evolutionary and ecological questions. From an ecological perspective, do the evolved differences in the two regional groups encompass significant differences in their biologies, and what mechanisms maintain this geographically sharp divide? We suggest that there are potentially two mechanisms that interlink: (a) restricted straying and/or gene flow, (b) divergent selective forces.
There are still many unknowns regarding straying rates among salmon populations, though they clearly vary in time and space Pedersen, Rasmussen, Nielsen, Karlsson, & Nyberg, 2007;Skilbrei & Holm, 1998;Stabell, 1984), and while some knowledge has been gained in recent years on their marine migration behavior (Chittenden, Adlandsvik, Pedersen, Righton, & Rikardsen, 2013;Gilbey et al., 2017;Gudjonsson, Einarsson, Jonsson, & Gudbrandsson, 2015;Strøm, Thorstad, Hedger, & Rikardsen, 2018), a large number of questions remain with respect to their oceanic migration routes and offshore feeding areas. Nevertheless, a lack of synchrony in marine growth of salmon populations from northern versus western Norway suggest that salmon originating from these two regions may utilize different oceanic feeding areas (Jensen et al., 2011). If this is the case, then fish retuning to the coastline in the region just north and south of the geographically limited transition zone identified here may come from different directions/oceanic areas, and act to reduce straying between the two regions. In turn, this could limit gene flow. However, a study of straying from two populations north of this transition zone found that fish strayed into rivers south of it (Ulvan et al., 2018), suggesting that the occurrence of some genetic mixing cannot be ruled out.
We observed a decrease in several estimators of genetic diversity with an increase in latitude ( Figure 2a) and a clear "shift" in allelic variation at two of the genetic markers in the transition zone where the aforementioned lineages meet (Figure 2b). In Canadian Atlantic salmon populations, a gradient in genetic diversity and allelic variation at the MHC2 locus has been reported (Dionne, Miller, Dodson, Caron, & Bernatchez, 2007), and a relationship between allele frequencies and latitude was observed for immune-related genes among European Atlantic salmon populations (Tonteri, Vasemägi, Lumme, & Primmer, 2010). Furthermore, allelic gradients with latitude and temperature have also been observed in respect of allelic variation at the MEP-2* locus on both sides of the Atlantic both within and among rivers (Verspoor & Jordan, 1989). A recent study using whole genome resequencing identified functional genetic differences between salmon populations from the north and the rest of Norway (Kjaerner-Semb et al., 2016), with evidence of islands of divergence on chromosomes 5, 10, 11, 13-15, 21, 24, and 25, possibly resulting from divergent selection regimes. This divergence included 59 known genes, 15 of which displayed one or more differentiated missense mutations. The strongest of these islands of divergence, located on chromosomes 25 and 5, respectively, contained genes involved in anti-viral and pathogen control. It is not possible to conclude the functional significance of the clear general decrease in genetic diversity as revealed in the present study, or specifically for two of the markers across the observed transition zone. While clearly further work is needed, what evidence there is points to the possibility of functional genetic differences between populations in these two regions, possibly arising from a combination of differences relating to phylogenetic background and lineage recolonization, and divergent selection regimes. If as suggested, divergent selection regimes between these areas exist, even if some interbreeding does occur due to straying across the transition zone, reduced survival of the "nonlocal" type, as observed across watercourses in Ireland (McGinnity et al., 2004), may strongly constrain effective gene flow and help maintain geographically restricted transition zone.

| Patterns of population genetic connectivity
A hierarchical pattern in genetic structure as revealed here, that is, within and among-regional levels of differentiation (Figures 3 and   4), also characterized by an overall pattern of isolation by distance (IBD) (Figure 5), is a typical feature of Atlantic salmon populations (Glover et al., 2012;Tonteri et al., 2009;Vaha, Erkinaro, Falkegard, Orell, & Niemela, 2017). In addition to the highest level regional differentiation in northern Norway, a further less marked splitting of the Barents-White Sea and eastern Atlantic lineages and several other genetic sub-groups was resolved (Figure 2). A second order division in population structure was reported in the Kola Peninsula of Russia between samples from the Ponoi and Iokanga (populations 9 and 10 in Figure 3). This corresponds to the division reported in earlier studies (Ozerov et al., 2017;Saisa et al., 2005;Tonteri et al., 2009) and to changes in the life-history pattern of populations (Berg, 1948). Other genetic divisions were also revealed (Figure 3), illustrating the existence of both long-distance and regional levels of genetic structure.
Population size differences (as evaluated from catch statistics or conservation limits), and potentially life history or other adaptive characteristics, appear linked with some of the patterns of genetic structure observed here. For example, in the relatively isolated Trondheimsfjord in mid-Norway, the ten rivers sampled show a clear pattern of genetic divergence between the rivers with demographically small populations (populations 66, 68, 70, 73, 74) and those with demographically large or very large populations (populations 67, 69, 71, 72, 75) (Figure 3). This effect is also apparent in respect of the rivers Gaula and Orkla, which are genetically very similar to each other (populations 72 and 75), yet very distinct from the two small populations located between them, Vigda and Børsa, that are also similar to each other (populations 73 and 74) (Figure 3). It is thus striking that the two very large rivers have not dominated or overridden the genetic characteristics of these two much smaller populations, something observed in studies in other regions (Verspoor, 2005;Verspoor, Knox, & Marshall, 2016), once again suggesting a role for adaptive divergence even on a local scale.
Landscape features are known to influence population genetic structure in Atlantic salmon (Dillane et al., 2008;Ozerov, Veselov, Lumme, & Primmer, 2012). Although beyond the scope of this study, obvious landscape features also appeared to be linked with some of the population genetic structure revealed here. For example, on the coastline of Jaeren on southwestern Norway, a genetic divide was revealed among populations in the Boknafjord region (populations 100-104) versus the immediately neighboring open coastline stretch of Jaeren (populations 105-109) (Figure 3). Also, the rivers located in the relatively isolated Trondheimsfjord area showed differentiation to rivers on the outside of this fjord area, which overlays the significant observations within the fjord as discussed above (Figure 3). Thus, there is considerable evidence that the evolutionary relationships among populations and their genetic differentiation is driven by more than just historical and contemporary gene flow conditioned by IBD.

ACK N OWLED G M ENTS
The authors wish to thank the numerous people who provided samples for genetic analysis, especially Rådgivende Biologer AS, who provided many of the samples from rivers in western Norway, the Norwegian Institute of Nature research, and the staff at the Freshwater Laboratory of PINRO-Russia, who collected samples from the rivers in Kola Peninsula. We are also grateful to Laila Unneland, Bjørghild Breistein Seliussen, and Anne Grete Sørvik Eide, who genotyped most of the samples.

CO N FLI C T O F I NTE R E S T S
The authors declare that they have no competing interests.

DATA ACCE SS I B I LIT Y
Appendices S1-S3 with data and supporting information, including the entire raw data set of individual genetic profiles for >9,000